[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-headwisekv-training-free-kv-cache-compression":3,"topics-all":31,"news-related-bc642ebf-9b1c-41cf-98d1-dd55393fd429":50},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":23,"news_slug":24,"published_at":25,"created_at":26,"modified_at":27,"is_published":28,"publish_type":29,"image_url":7,"view_count":30},"bc642ebf-9b1c-41cf-98d1-dd55393fd429","HeadWiseKV:无训练KV cache压缩让混合LLM长上下文从114K推到161K","长上下文推理的痛点之一是KV cache随上下文增长,持续抢占GPU显存并压低吞吐。这个问题在Qwen3.8-Flash-Next、Qwen3.8-Max这类混合架构模型上更突出——它们把局部注意力、循环模块、线性注意力与少量全局注意力层叠加,而正是那几层残余的全局注意力决定了cache大小,也成了吞吐的瓶颈。\\n\\nRenjie Xie等人9月2日提交的HeadWiseKV(arXiv:2609.02029)给出训练免费的解法。它的核心是为每个物理KV head预先绑定一个静态的多级历史窗口,让cache占用在服务启动前就可预测;并把这个分配建模成受限的率失真问题,提出SeqCalib算法按层序生成策略,显式考虑层间policy的相互作用。最后用一个grouped-cache运行时把策略落地成真实的per-head驻留,而不是给full cache打mask。\\n\\n论文在Qwen3.6-27B等四个混合模型上验证:HeadWiseKV把112K上下文的峰值显存降低8.59%,并把「能稳定跑通」最大上下文从114K推到161K(约+41%),同时在RULER和LoCoMo长上下文基准上几乎不掉点。最关键的是它训练免费——存量混合架构模型直接挂载即可受益,不必重训或微调。\\n\\n对运营长上下文推理服务的人来说,这类工作正在把「显存墙」从结构性瓶颈变成调优问题。值得继续关注的是它与UltraQuant等KV cache量化方法的组合空间,以及在RAG、Agent长轨迹等真实场景里的端到端收益。",null,"https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.02029","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,14,17,20],{"id":12,"name":13,"slug":13,"description":7,"color":7},"fca9258a-9430-455a-b95d-b9fae5e373a8","ai-inference",{"id":15,"name":16,"slug":16,"description":7,"color":7},"2d9c2fb0-2be5-4ad1-aedb-e9747addf355","compression",{"id":18,"name":19,"slug":19,"description":7,"color":7},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",{"id":21,"name":22,"slug":22,"description":7,"color":7},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[],"headwisekv-training-free-kv-cache-compression","2026-09-03T03:44:00Z","2026-09-03T03:49:37.156897Z","2026-09-03T03:49:37.156908Z",true,"agent",194,[32,41],{"slug":33,"tag_slug":33,"title_zh":34,"title_en":35,"intro_zh":36,"intro_en":37,"id":38,"is_active":28,"created_at":39,"modified_at":40},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":42,"tag_slug":42,"title_zh":43,"title_en":44,"intro_zh":45,"intro_en":46,"id":47,"is_active":28,"created_at":48,"modified_at":49},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":51},[52,57,62,67,72,77],{"id":53,"title":54,"news_slug":55,"published_at":56},"47f025dd-69eb-4fcb-b4cc-1c6d4e03ca66","PartInfer：神经元级优化突破边缘设备LLM推理瓶颈","partinfer-neuron-level-edge-llm","2026-06-03T04:00:00+00:00",{"id":58,"title":59,"news_slug":60,"published_at":61},"e91b3add-4f1d-48d3-ab7a-bd2c6e8c1765","QuIP 崩、OPTQ 降级:Kashin-DCT 在 4-bit 量化压力测试里活了下来","kashin-dct-2bit-llm-quantization","2026-09-12T15:10:00+00:00",{"id":63,"title":64,"news_slug":65,"published_at":66},"c83af54b-79ed-445c-9482-07d98c26c36b","BeaconKV:长推理会回头看,只压最近窗口的 KV 缓存注定丢东西","beaconkv-beacon-query-kv-cache-compression","2026-09-09T11:25:00+00:00",{"id":68,"title":69,"news_slug":70,"published_at":71},"bce9fc16-d31a-49be-b17b-f144619a58e2","LatentPress:上下文压成软令牌直读，7.7 倍压缩反超原文，训练仅动 0.1% 参数","latentpress-soft-token-context-compression","2026-09-05T19:06:09+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"d31bc388-b6c7-41a5-a6e9-6f00657c7616","加GPU还是压KV缓存？arXiv论文：压缩省钱1.2到2倍，但36B是道坎","tensor-parallelism-vs-kv-compression-cost","2026-08-30T17:10:00+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"f65e204c-0115-4b50-9113-2c3bb2ff6637","ReCache:给 Agent 的工具记忆装上独立缓存,KV 内存砍 92%、首 token 提速 3.655 倍","recache-agent-kv-cache-reuse","2026-08-24T15:30:00+00:00"]